DocuChat is a Next.js app for document-grounded chat. Upload files, index them in Pinecone, and ask questions against retrieved context.

• Upload flow for PDF, DOCX, TXT, MD, and image files (PNG/JPG/JPEG)
• Text extraction pipeline:
  • PDF via pdf-parse
  • DOCX via mammoth
  • Images via tesseract.js OCR
  • TXT/MD as UTF-8 text
• Chunking and embedding pipeline with LangChain
• Pinecone-backed vector retrieval
• Chat endpoint grounded in retrieved chunks with source metadata
• Sources panel in the UI showing file + chunk provenance
• BYOK (Bring Your Own Key) in the UI via local storage

• Next.js (App Router) + React + TypeScript
• LangChain
• Pinecone
• OpenAI chat + embeddings
• Tailwind CSS + Framer Motion

• Node.js 20+
• A Pinecone project and index
• OpenAI API key (server key and/or BYOK key in the UI)

Create .env.local in the project root:

PINECONE_API_KEY=your_pinecone_api_key
PINECONE_INDEX=docuchat-index
PINECONE_NAMESPACE=optional_namespace

OPENAI_API_KEY=your_openai_api_key
OPENAI_CHAT_MODEL=gpt-4o-mini
OPENAI_EMBEDDING_MODEL=text-embedding-3-small
OPENAI_EMBEDDING_DIMENSION=1536

Notes:

• PINECONE_API_KEY is required.
• If OPENAI_API_KEY is not set on the server, users must set a BYOK key in the app before upload/chat will work.
• OPENAI_EMBEDDING_DIMENSION should match your Pinecone index dimension when needed.

npm install
npm run dev

Open http://localhost:3000.

1. Open the Upload view and import a document.
2. Wait for processing + indexing to complete.
3. Open Chat and ask questions about the uploaded content.
4. Optional: use the BYOK button in the header to store your OpenAI key in your browser.

• npm run dev - start development server
• npm run build - create production build
• npm run start - run production server
• npm run lint - run ESLint